2006-01-10 Frank Ch. Eigler <fche@redhat.com>

	PR 2060.
	* buildrun.cxx (compile_pass): Add "V=1" to kbuild if verbose.
	* translate.cxx (translator_output): For output-file constructor,
	set an explicit output buffer.
	(emit_module_init, emit_module_exit): Reorganize output, to spit
	each individual probe registration/deregistration blurb into a
	separate function.
	* translate.h: Corresponding changes; set default buffer size to 8K.
	* translate.cxx, tapsets.cxx: Replace "endl" by buffer-friendly "\n"
	throughout code generation routines.

5 files changed, 125 insertions, 81 deletions

diff --git a/ChangeLog b/ChangeLog
index ebfa15c6..1178c456 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,16 @@
+2006-01-10  Frank Ch. Eigler  <fche@redhat.com>
+
+	PR 2060.
+	* buildrun.cxx (compile_pass): Add "V=1" to kbuild if verbose.
+	* translate.cxx (translator_output): For output-file constructor,
+	set an explicit output buffer.
+	(emit_module_init, emit_module_exit): Reorganize output, to spit
+	each individual probe registration/deregistration blurb into a
+	separate function.
+	* translate.h: Corresponding changes; set default buffer size to 8K.
+	* translate.cxx, tapsets.cxx: Replace "endl" by buffer-friendly "\n"
+	throughout code generation routines.
+
 2006-01-09  Frank Ch. Eigler  <fche@redhat.com>
 
 	* HACKING: Extend guidelines for tapset testing.
diff --git a/buildrun.cxx b/buildrun.cxx
index 05d903d5..d678e874 100644
--- a/buildrun.cxx
+++ b/buildrun.cxx
@@ -66,6 +66,9 @@ compile_pass (systemtap_session& s)
   for (unsigned i=0; i<s.macros.size(); i++)
     o << "CFLAGS += -D " << lex_cast_qstring(s.macros[i]) << endl;
 
+  // XXX
+  // o << "CFLAGS += -ftime-report" << endl;
+
   if (s.test_mode)
     {
       string module_dir = string("/lib/modules/")
@@ -95,7 +98,9 @@ compile_pass (systemtap_session& s)
     {
       string make_cmd = string("make -C \"") + s.tmpdir + "\"";
 
-      if (! s.verbose)
+      if (s.verbose)
+        make_cmd += " V=1";
+      else
         make_cmd += " -s >/dev/null 2>&1";
       
       if (s.verbose) clog << "Running " << make_cmd << endl;
@@ -113,7 +118,9 @@ compile_pass (systemtap_session& s)
         + string (" -C \"") + module_dir + string("\"");
       make_cmd += string(" M=\"") + s.tmpdir + string("\" modules");
 
-      if (! s.verbose)
+      if (s.verbose)
+        make_cmd += " V=1";
+      else
         make_cmd += " -s >/dev/null 2>&1";
       
       if (s.verbose) clog << "Running " << make_cmd << endl;
diff --git a/tapsets.cxx b/tapsets.cxx
index b6533a10..57429cbd 100644
--- a/tapsets.cxx
+++ b/tapsets.cxx
@@ -217,7 +217,7 @@ be_derived_probe::emit_probe_entries (translator_output* o, unsigned j)
 
       emit_probe_epilogue (o);
 
-      o->newline(-1) << "}" << endl;
+      o->newline(-1) << "}\n";
     }
 }
 
@@ -456,7 +456,7 @@ dwflpp
     if (in)
       return in;
     if (false && sess.verbose)
-      clog << "WARNING: no name found for " << type << endl;
+      clog << "WARNING: no name found for " << type << "\n";
     return string("");
   }
 
@@ -482,7 +482,7 @@ dwflpp
 	if (required)
 	  throw semantic_error (msg);
 	else
-	  cerr << "WARNING: " << msg << endl;
+	  cerr << "WARNING: " << msg << "\n";
       }
   }
 
@@ -549,8 +549,6 @@ dwflpp
   {
     assert(dwfl);
     cu = NULL;
-    if (false && sess.verbose)
-      clog << "focusing on module containing global addr " << a << endl;
     Dwfl_Module* mod = dwfl_addrmodule(dwfl, a);
     if (mod) // address could be wildly out of range
       focus_on_module(mod);
@@ -586,7 +584,7 @@ dwflpp
     if (false && sess.verbose)
       clog << "module addr " << hex << a
 	   << " + module start " << module_start
-	   << " -> global addr " << (a + module_start) << dec << endl;
+	   << " -> global addr " << (a + module_start) << dec << "\n";
     return a + module_start;
   }
 
@@ -598,7 +596,7 @@ dwflpp
     if (false && sess.verbose)
       clog << "global addr " << a
 	   << " - module start " << hex << module_start
-	   << " -> module addr " << (a - module_start) << dec << endl;
+	   << " -> module addr " << (a - module_start) << dec << "\n";
     return a - module_bias;
   }
 
@@ -610,7 +608,7 @@ dwflpp
     if (t && sess.verbose)
       clog << "pattern '" << pattern << "' "
 	   << "matches "
-	   << "module '" << module_name << "'" << endl;
+	   << "module '" << module_name << "'" << "\n";
     return t;
   }
 
@@ -622,7 +620,7 @@ dwflpp
     if (t && sess.verbose)
       clog << "pattern '" << pattern << "' "
 	   << "matches "
-	   << "function '" << function_name << "'" << endl;
+	   << "function '" << function_name << "'" << "\n";
     return t;
   }
 
@@ -634,7 +632,7 @@ dwflpp
     if (t && sess.verbose)
       clog << "pattern '" << pattern << "' "
 	   << "matches "
-	   << "CU '" << cu_name << "'" << endl;
+	   << "CU '" << cu_name << "'" << "\n";
     return t;
   }
 
@@ -893,7 +891,7 @@ dwflpp
 	  {
 	    filtered_srcfiles.insert (fname);
 	    if (sess.verbose)
-	      clog << "selected source file '" << fname << "'" << endl;
+	      clog << "selected source file '" << fname << "'\n";
 	  }
       }
     }
@@ -994,7 +992,7 @@ dwflpp
                        << last_function->name
 		       << " heur0=" << hex << addr0
 		       << " heur1=" << addr << dec
-		       << endl;
+		       << "\n";
               }
 	    choose_next_line_otherthan = -1;
             continue;
@@ -1017,7 +1015,7 @@ dwflpp
 		  clog << "prologue disagreement: " << last_function->name
 		       << " heur0=" << hex << addr0
 		       << " heur1=" << addr << dec
-		       << endl;
+		       << "\n";
               }
 	    choose_next_line_otherthan = -1;
             continue;
@@ -1043,7 +1041,7 @@ dwflpp
                    << last_function->name
                    << "' entrypc=0x" << hex << addr 
                    << " highpc=0x" << last_function_highpc
-                   << endl;
+                   << "\n";
           }
       }
 
@@ -1437,7 +1435,7 @@ dwflpp
       clog << "finding location for local '" << local
 	   << "' near address " << hex << pc
 	   << ", module bias " << module_bias << dec
-	   << endl;
+	   << "\n";
 
     Dwarf_Attribute attr_mem;
     if (dwarf_attr_integrate (&vardie, DW_AT_location, &attr_mem) == NULL)
@@ -1794,7 +1792,7 @@ dwarf_query::parse_function_spec(string & spec)
       if (sess.verbose)
 	clog << "parsed '" << spec
 	     << "' -> func '" << function
-	     << "'" << endl;
+	     << "'\n";
       return function_alone;
     }
 
@@ -1810,7 +1808,7 @@ dwarf_query::parse_function_spec(string & spec)
 	clog << "parsed '" << spec
 	     << "' -> func '"<< function
 	     << "', file '" << file
-	     << "'" << endl;
+	     << "'\n";
       return function_and_file;
     }
 
@@ -1824,7 +1822,7 @@ dwarf_query::parse_function_spec(string & spec)
 	clog << "parsed '" << spec
 	     << "' -> func '"<< function
 	     << "', file '" << file
-	     << "', line " << line << endl;
+	     << "', line " << line << "\n";
       return function_file_and_line;
     }
   catch (runtime_error & exn)
@@ -1935,7 +1933,7 @@ dwarf_query::blacklisted_p(string const & funcname,
 	  if (sess.verbose)
 	    clog << "skipping function '" << funcname << "' base 0x"
 		 << hex << addr << dec << " is within section '"
-		 << name << "'" << endl;
+		 << name << "'\n";
 	  return true;
 	}
     }
@@ -1970,7 +1968,7 @@ dwarf_query::blacklisted_p(string const & funcname,
 		  if (sess.verbose)
 		    clog << "skipping function '" << funcname << "' base 0x"
 			 << hex << addr << dec << " is within section '"
-			 << name << "'" << endl;
+			 << name << "'\n";
 		  return true;
 		}
 	    }
@@ -1993,7 +1991,7 @@ dwarf_query::blacklisted_p(string const & funcname,
     {
       if (sess.verbose)
         clog << "skipping function '" << funcname << "' file '"
-             << filename << "' is blacklisted" << endl;
+             << filename << "' is blacklisted\n";
       return true;
     }
 
@@ -2113,7 +2111,7 @@ query_inline_instance_info (Dwarf_Addr entrypc,
 	  if (q->sess.verbose)
 	    clog << "querying entrypc "
 		 << hex << entrypc << dec
-		 << " of instance of inline '" << ii.name << "'" << endl;
+		 << " of instance of inline '" << ii.name << "'\n";
 	  query_statement (ii.name, ii.decl_file, ii.decl_line,
 			   &ii.die, entrypc, q);
 	}
@@ -2176,7 +2174,7 @@ query_srcfile_line (Dwarf_Line * line, void * arg)
       if (q->dw.die_has_pc (&(i->second.die), addr))
 	{
 	  if (q->sess.verbose)
-	    clog << "function DIE lands on srcfile" << endl;
+	    clog << "function DIE lands on srcfile\n";
 	  if (q->has_statement_str)
 	    query_statement (i->second.name, i->second.decl_file,
 			     q->line, NULL, addr, q);
@@ -2192,7 +2190,7 @@ query_srcfile_line (Dwarf_Line * line, void * arg)
       if (q->dw.die_has_pc (&(i->second.die), addr))
 	{
 	  if (q->sess.verbose)
-	    clog << "inline instance DIE lands on srcfile" << endl;
+	    clog << "inline instance DIE lands on srcfile\n";
 	  if (q->has_statement_str)
 	    query_statement (i->second.name, i->second.decl_file,
 			     q->line, NULL, addr, q);
@@ -2215,7 +2213,7 @@ query_dwarf_inline_instance (Dwarf_Die * die, void * arg)
       bool record_this_inline = false;
 
       if (q->sess.verbose)
-	clog << "examining inline instance of " << q->dw.function_name << endl;
+	clog << "examining inline instance of " << q->dw.function_name << "\n";
 
       if (q->has_inline_str || q->has_statement_str)
 	record_this_inline = true;
@@ -2234,7 +2232,8 @@ query_dwarf_inline_instance (Dwarf_Die * die, void * arg)
       if (record_this_inline)
 	{
 	  if (q->sess.verbose)
-	    clog << "selected inline instance of " << q->dw.function_name << endl;
+	    clog << "selected inline instance of " << q->dw.function_name
+                 << "\n";
 
 	  Dwarf_Addr entrypc;
 	  if (q->dw.die_entrypc (die, &entrypc))
@@ -2281,7 +2280,8 @@ query_dwarf_func (Dwarf_Die * func, void * arg)
 	      || q->has_inline_num))
 	{
 	  if (q->sess.verbose)
-	    clog << "checking instances of inline " << q->dw.function_name << endl;
+	    clog << "checking instances of inline " << q->dw.function_name
+                 << "\n";
 	  q->dw.iterate_over_inline_instances (query_dwarf_inline_instance, arg);
 	}
       else if (!q->dw.func_is_inline ())
@@ -2310,7 +2310,7 @@ query_dwarf_func (Dwarf_Die * func, void * arg)
 	  if (record_this_function)
 	    {
 	      if (q->sess.verbose)
-		clog << "selected function " << q->dw.function_name << endl;
+		clog << "selected function " << q->dw.function_name << "\n";
 
 	      Dwarf_Addr entrypc;
 	      if (q->dw.function_entrypc (&entrypc))
@@ -2347,7 +2347,7 @@ query_cu (Dwarf_Die * cudie, void * arg)
 
       if (false && q->sess.verbose)
         clog << "focused on CU '" << q->dw.cu_name
-             << "', in module '" << q->dw.module_name << "'" << endl;
+             << "', in module '" << q->dw.module_name << "'\n";
 
       if (q->has_statement_str
 	  || q->has_inline_str || q->has_inline_num
@@ -2474,7 +2474,7 @@ query_module (Dwfl_Module *mod __attribute__ ((unused)),
 	clog << "focused on module '" << q->dw.module_name
 	     << "' = [" << hex << q->dw.module_start
 	     << "-" << q->dw.module_end
-	     << ", bias " << q->dw.module_bias << "]" << dec << endl;
+	     << ", bias " << q->dw.module_bias << "]" << dec << "\n";
 
       if (q->has_inline_num || q->has_function_num || q->has_statement_num)
         {
@@ -3121,7 +3121,7 @@ dwarf_derived_probe::emit_probe_entries (translator_output* o,
   emit_probe_epilogue (o);
 
   o->newline() << "return 0;";
-  o->newline(-1) << "}" << endl;
+  o->newline(-1) << "}\n";
   if (has_return)
     o->newline() << "#endif /* ARCH_SUPPORTS_KRETPROBES */";
 
@@ -3291,7 +3291,7 @@ timer_derived_probe::emit_probe_entries (translator_output* o, unsigned j)
   o->newline() << "probe_" << j << " (c);";
 
   emit_probe_epilogue (o);
-  o->newline(-1) << "}" << endl;
+  o->newline(-1) << "}\n";
 }
 
 
@@ -3410,7 +3410,7 @@ profile_derived_probe::emit_probe_entries (translator_output* o, unsigned j)
 
   emit_probe_epilogue (o);
   o->newline() << "return 0;";
-  o->newline(-1) << "}" << endl;
+  o->newline(-1) << "}\n";
 }
 
 
diff --git a/translate.cxx b/translate.cxx
index 791e5434..992bd96c 100644
--- a/translate.cxx
+++ b/translate.cxx
@@ -776,20 +776,25 @@ ostream & operator<<(ostream & o, itervar const & v)
 
 
 translator_output::translator_output (ostream& f):
-  o2 (0), o (f), tablevel (0)
+  buf(0), o2 (0), o (f), tablevel (0)
 {
 }
 
 
-translator_output::translator_output (const string& filename):
-  o2 (new ofstream (filename.c_str ())), o (*o2), tablevel (0)
+translator_output::translator_output (const string& filename, size_t bufsize):
+  buf (new char[bufsize]),
+  o2 (new ofstream (filename.c_str ())), 
+  o (*o2), 
+  tablevel (0)
 {
+  o2->rdbuf()->pubsetbuf(buf, bufsize);
 }
 
 
 translator_output::~translator_output ()
 {
   delete o2;
+  delete buf;
 }
 
 
@@ -798,7 +803,7 @@ translator_output::newline (int indent)
 {
   assert (indent > 0 || tablevel >= (unsigned)-indent);
   tablevel += indent;
-  o << endl;
+  o << "\n";
   for (unsigned i=0; i<tablevel; i++)
     o << "  ";
   return o;
@@ -921,13 +926,13 @@ c_unparser::emit_common_header ()
       o->newline(-1) << "} function_" << c_varname (fd->name) << ";";
     }
   o->newline(-1) << "} locals [MAXNESTING];";
-  o->newline(-1) << "};" << endl;
-  o->newline() << "void *contexts = NULL; /* alloc_percpu */" << endl;
+  o->newline(-1) << "};\n";
+  o->newline() << "void *contexts = NULL; /* alloc_percpu */\n";
 
   emit_map_type_instantiations ();
 
   if (!session->stat_decls.empty())
-    o->newline() << "#include \"stat.c\"" << endl;
+    o->newline() << "#include \"stat.c\"\n";
 }
 
 
@@ -966,8 +971,45 @@ c_unparser::emit_functionsig (functiondecl* v)
 void
 c_unparser::emit_module_init ()
 {
-  o->newline() << "static int systemtap_module_init (void);";
-  o->newline() << "int systemtap_module_init () {";
+  // Emit the per-probe-point registrations into individual functions,
+  // to avoid forcing the compiler to work too hard at optimizing such
+  // a silly function.  A "don't optimize this function" pragma could
+  // come in handy too.
+  for (unsigned i=0; i<session->probes.size(); i++)
+    {
+      o->newline() << "noinline int register_probe_" << i << " (void) {";
+      o->indent(1);
+      // By default, mark the first location as the site of possible
+      // registration failure.  This is helpful since non-dwarf
+      // derived_probes tend to have only a single location.
+      assert (session->probes[i]->locations.size() > 0);
+      o->newline() << "int rc = 0;";
+      o->newline() << "const char *probe_point = " <<
+        lex_cast_qstring (*session->probes[i]->locations[0]) << ";";
+      session->probes[i]->emit_registrations (o, i);
+
+      o->newline() << "if (unlikely (rc)) {";
+      // In case it's just a lower-layer (kprobes) error that set rc
+      // but not session_state, do that here to prevent any other BEGIN
+      // probe from attempting to run.
+      o->newline(1) << "atomic_set (&session_state, STAP_SESSION_ERROR);";
+
+      o->newline() << "_stp_error (\"probe " << i << " registration failed"
+                   << ", rc=%d, %s\\n\", rc, probe_point);";
+      o->newline(-1) << "}";
+
+      o->newline() << "return rc;";
+      o->newline(-1) << "}";
+      
+      o->newline();
+      o->newline() << "noinline void unregister_probe_" << i << " (void) {";
+      o->indent(1);
+      session->probes[i]->emit_deregistrations (o, i);
+      o->newline(-1) << "}";
+    }
+
+  o->newline();
+  o->newline() << "int systemtap_module_init (void) {";
   o->newline(1) << "int rc = 0;";
   o->newline() << "const char *probe_point = \"\";";
 
@@ -998,34 +1040,16 @@ c_unparser::emit_module_init ()
 
   for (unsigned i=0; i<session->probes.size(); i++)
     {
-      o->newline() << "/* register probe #" << i << ", ";
-      o->line() << session->probes[i]->locations.size() << " location(s) */";
-
-      // By default, mark the first location as the site of possible
-      // registration failure.  This is helpful since non-dwarf
-      // derived_probes tend to have only a single location.
-      assert (session->probes[i]->locations.size() > 0);
-      o->newline() << "probe_point = " <<
-        lex_cast_qstring (*session->probes[i]->locations[0]) << ";";
-      session->probes[i]->emit_registrations (o, i);
-
-      o->newline() << "if (unlikely (rc)) {";
-      // In case it's just a lower-layer (kprobes) error that set rc
-      // but not session_state, do that here to prevent any other BEGIN
-      // probe from attempting to run.
-      o->newline(1) << "atomic_set (&session_state, STAP_SESSION_ERROR);";
-
-      o->newline() << "_stp_error (\"probe " << i << " registration failed"
-                   << ", rc=%d, %s\\n\", rc, probe_point);";
-
+      o->newline() << "rc = register_probe_" << i << "();";
+      o->newline() << "if (rc)";
+      o->indent(1);
       // We need to deregister any already probes set up - this is
       // essential for kprobes.
       if (i > 0)
         o->newline() << "goto unregister_" << (i-1) << ";";
       else
         o->newline() << "goto out;";
-
-      o->newline(-1) << "}";
+      o->indent(-1);
     }
 
   // BEGIN probes would have all been run by now.  One of them may
@@ -1045,8 +1069,7 @@ c_unparser::emit_module_init ()
   for (int i=session->probes.size()-2; i >= 0; i--) // NB: -2
     {
       o->newline(-1) << "unregister_" << i << ":";
-      o->indent(1);
-      session->probes[i]->emit_deregistrations (o, i);
+      o->newline(1) << "unregister_probe_" << i << "();";
       // NB: This may be an END probe.  It will refuse to run
       // if the session_state was ERRORed.
     }  
@@ -1063,15 +1086,14 @@ c_unparser::emit_module_init ()
 
   o->newline(-1) << "out:";
   o->newline(1) << "return rc;";
-  o->newline(-1) << "}" << endl;
+  o->newline(-1) << "}\n";
 }
 
 
 void
 c_unparser::emit_module_exit ()
 {
-  o->newline() << "static void systemtap_module_exit (void);";
-  o->newline() << "void systemtap_module_exit () {";
+  o->newline() << "void systemtap_module_exit (void) {";
   // rc?
   o->newline(1) << "int holdon;";
 
@@ -1107,7 +1129,7 @@ c_unparser::emit_module_exit ()
   // genuinely stuck somehow
 
   for (int i=session->probes.size()-1; i>=0; i--)
-    session->probes[i]->emit_deregistrations (o, i); // NB: runs "end" probes
+    o->newline() << "unregister_probe_" << i << "();"; // NB: runs "end" probes
 
   for (unsigned i=0; i<session->globals.size(); i++)
     {
@@ -1127,7 +1149,7 @@ c_unparser::emit_module_exit ()
                 << "(int) atomic_read (& skipped_count));";
   o->indent(-1);
 
-  o->newline(-1) << "}" << endl;
+  o->newline(-1) << "}\n";
 }
 
 
@@ -1179,7 +1201,7 @@ c_unparser::emit_function (functiondecl* v)
 
   o->newline() << "#undef CONTEXT";
   o->newline() << "#undef THIS";
-  o->newline(-1) << "}" << endl;
+  o->newline(-1) << "}\n";
 }
 
 
@@ -1187,7 +1209,7 @@ void
 c_unparser::emit_probe (derived_probe* v, unsigned i)
 {
   // o->newline() << "static void probe_" << i << " (struct context *c);";
-  o->newline() << "static void probe_" << i << " (struct context * __restrict__ c) {";
+  o->newline() << "void probe_" << i << " (struct context * __restrict__ c) {";
   o->indent(1);
 
   // initialize frame pointer
@@ -1222,7 +1244,8 @@ c_unparser::emit_probe (derived_probe* v, unsigned i)
   o->newline(-1) << "out:";
   // NB: no need to uninitialize locals, except if arrays can somedays be local
   o->newline(1) << "_stp_print_flush();";
-  o->newline(-1) << "}" << endl;
+
+  o->newline(-1) << "}\n";
   
   v->emit_probe_entries (o, i);
 }
@@ -3455,7 +3478,7 @@ emit_symbol_data (systemtap_session& s)
 	    }
 	}
       s.op->newline(-1) << "};";
-      s.op->newline() << "unsigned stap_num_symbols = " << i << ";" << endl;
+      s.op->newline() << "unsigned stap_num_symbols = " << i << ";\n";
     }
 
   return rc;
@@ -3474,7 +3497,7 @@ translate_pass (systemtap_session& s)
   try
     {
       // This is at the very top of the file.
-      s.op->line() << "#define TEST_MODE " << (s.test_mode ? 1 : 0) << endl;
+      s.op->line() << "#define TEST_MODE " << (s.test_mode ? 1 : 0) << "\n";
 
       s.op->newline() << "#ifndef MAXNESTING";
       s.op->newline() << "#define MAXNESTING 10";
@@ -3526,7 +3549,7 @@ translate_pass (systemtap_session& s)
 
       for (unsigned i=0; i<s.embeds.size(); i++)
         {
-          s.op->newline() << s.embeds[i]->code << endl;
+          s.op->newline() << s.embeds[i]->code << "\n";
         }
 
       for (unsigned i=0; i<s.globals.size(); i++)
@@ -3590,7 +3613,7 @@ translate_pass (systemtap_session& s)
     }
 
   rc |= emit_symbol_data (s);
-  s.op->line() << endl;
+  s.op->line() << "\n";
 
   delete s.op;
   s.op = 0;
diff --git a/translate.h b/translate.h
index 39e56dd7..1c6dc9f6 100644
--- a/translate.h
+++ b/translate.h
@@ -21,13 +21,14 @@
 // pretty-printing.
 class translator_output
 {
+  char *buf;
   std::ofstream* o2;
   std::ostream& o;
   unsigned tablevel;
 
 public:
   translator_output (std::ostream& file);
-  translator_output (const std::string& filename);
+  translator_output (const std::string& filename, size_t bufsize = 8192);
   ~translator_output ();
 
   std::ostream& newline (int indent = 0);